Characterising ambient air quality over Gauteng using surface in-situ and remotely derived measurements

Abstract

Air quality studies using ground-based and satellite retrievals have gained popularity over the past two decades, offering many techniques and methods to explain atmospheric observations and contribute to climate predictions. Satellite retrievals assist us in capturing what ground-based measurements cannot; however, this is not to infer that one can be successfully employed as a proxy for the other. Ground-based measurements have presented major limitations on spatial coverage and data quality. This study aims to comprehensively assess air quality in the Gauteng Province, focusing on urban and industrial communities, by combining ground-based and remotely derived measurements. The first objective was to characterise ambient PM2.5 levels in the Gauteng region using ground-based measurements by identifying and analysing temporal trends, diurnal variations, and seasonal fluctuations in ground-based PM2.5 measurements. Secondary data for pollutant concentrations and meteorological data was used. The second objective is to evaluate the accuracy of satellite-derived data over the Gauteng region by validating the Aerosol Optical Depth (AOD) observations for MODISTERRA and MODISAQUA using the ground-based Aerosol Robotic Network (AERONET) observations. This objective used three retrieval algorithms, namely “Deep Blue (DB)”, “Dark Target (DT)”, and “Combined Deep Blue and Dark Target (DB & DT)”, to facilitate satellite AOD validations. The last objective was to characterise the aerosol loadings over Gauteng using satellite observations and modelled reanalysis data by comparing observations of AOD from MODISTERRA, MODISAQUA, and the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2) as well as characterising the aerosol type using the Ångström Exponent (α) for the monitored and unmonitored regions in Gauteng. ArcMap v10.8.1 was used for raster analysis and visualisation. The results show that PM2.5 concentrations are still localised in low-income communities where major sources of pollution are biomass and waste burning, and domestic fuel burning, and industrial activities. When validating AOD, the “Combined DB & DB” algorithm performed best with the Machine Learning (ML) – SciKit-Learn model with a 90% model performance as opposed to the SLR and OLS statistical models. The retrieval algorithm performed better with satellite data from MODISAQUA. MERRA-2 was the most efficient in capturing aerosol species distribution. The findings highlight the importance of satellite validation studies as meteorological and topographic features impact different geographic areas. The findings suggest exploring different atmospheric measurement and monitoring techniques to fully understand the limitations in air quality studies and finding measures fully capture the PM2.5 spatial distribution and AOD coverage.